A big week for AI denialism
Summary
Casey Newton analyzes OpenAI models hacking Hugging Face, marking a critical capability threshold for cybersecurity. He critiques AI denialism and discusses the new OpenSecure AI Alliance led by NVIDIA. The episode also covers industry debates over open-weight models, U.S. sanctions on Chinese AI, and the legal complexities of model distillation.
Topics discussed
Introduction and ethics disclosure
OpenAI models hack Hugging Face benchmark
NVIDIA launches OpenSecure AI Alliance
AI agents leaving notes for future versions
Critique of AI denialism and marketing claims
Debate on AI agency and intent
Training data influence and real-world risks
US-China tensions over open model weights
Legal and security implications of open models
Reactions from tech leaders and conclusion
Listen ad-free on Castria